Inspiration
Retail crypto investors lose money to FOMO: buying after vertical pumps, ignoring token-unlock pressure and hidden supply inflation. Most tools show one number or one opinion. We wanted a tool that shows what is measured, what is assumed, and what is missing — honesty over polish.
What it does
AuraMatrix AI pulls live market data from CoinGecko and computes real risk statistics (volatility, Sharpe ratio, max drawdown, Monte Carlo VaR/CVaR). It produces a transparent Aura-Score (0-100) across 9 components, where missing data can never raise the score. Three Meta Model API agents (Macro, On-chain, Quant risk) debate in parallel, a fourth writes the synthesis, and a Devil's Advocate agent tries to refute the score using only the analysis data. A What-if agent lets users ask "what if volume drops to 1M?" — the model calls the real scoring code as a tool instead of guessing.
How we built it
Built on a Streamlit prototype, we integrated Meta's Model API (muse-spark-1.3) for the multi-agent dialectic, added live CoinGecko data, real quantitative risk metrics, and a transparent 9-component scoring engine with an explicit uncertainty premium for missing data. Meta's Muse Code wrote the automated test suite (16 tests).
Challenges we ran into
Keeping the AI agents honest: making sure they never invent data when it's unavailable, and that no mock text is ever presented as real AI output. We also ran a first back-test and found the score doesn't beat a plain volatility baseline as a predictor — we reported that honestly instead of hiding it or overfitting the thresholds to hide it.
Accomplishments that we're proud of
A fully transparent scoring methodology where every component and penalty is documented, bilingual support (English/Turkish) from one dictionary, and real tool-calling where the AI calls our actual scoring code rather than guessing numbers.
What we learned
That rigorous validation (back-testing) matters more than a polished score — and that admitting a tool's limits honestly is more valuable than a confident but untested claim.
What's next for AuraMatrix AI
Adding on-chain whale/liquidation data once a reliable free source is available, and re-testing the scoring thresholds on a different time period before any recalibration.
